jamtholee / flags Goto Github PK

Any chance you could share the actual code to see what is going on under the hood?
Particularly wrt the NN model?

The neural network prediction does not seem to be working.

Our systems experienced an outage which started at 5:30am on December 2nd and lasted until 9:45 on December 5th. Data is currently missing for this period, but efforts will be made to restore it.

Note that this outage overlaps with the separate outage of the Osney lock stream gauges - See #5 for details.

We are slowly dealing with an ongoing server resource issue and this will result in occasional outages as we diagnose & solve the problem. Now that term has started and we get in to winter, we will try to limit website downtime to hours rather than days, and will try to restore any data not collected live.

Hi,

The current prediction is showing "Grey", which you said normally means you need to reboot a server.

Cheers!

The sensors at Osney lock appear to have another issue, which started at 7pm on December 3rd, and is affecting both the API and the EA website. As before, there is no estimate for when service will resume, since this is an issue coming from the EA data provider.

Note that this outage overlaps with the outage to our systems between December 2nd 5:30am and December 4th 9:45am - See #6 for details.

The sensors at Osney lock appeared to have an issue which started at 2pm on October 29th. Originally this affected both the API and the website, although the website came back online at some point on November 2nd. At time of writing, on November 3rd, the API is still not functioning normally, and the graphs are still not updating.

There is currently no estimate for when this will be back online, since it is an issue coming from the EA, but we will look into using the website as a backup live source, and ways to retrieve historical data from the missing period.

The new prediction tracking is really helpful, but it would be very instructive to have that plotted with the flow. Would this be possible to implement?

I assume the Grey flag is obtained from the OURCs flag status API, and is not predicted by your model.

Could you predict the flag colour just based on your model, and display the actual flag colour as a separate data field?

What causes a grey prediction here?

Any idea what's been causing the few weird readings recently?

• Thursday, Dec 01, 11:00, reading way too high
• Friday, Nov 25, 14:15, reading something big and negative
• Tuesday, Nov 22, immediately after 21:30, reading something big and negative

The main issue is auto scaling on the y-axis doesn't play well with the extremely high reading from a few days ago.

The Osney Downstream sensor in particular is a bit flakey, and sometimes goes offline, meaning the graphs stop showing any data at all (even if the Iffley upstream sensor is reporting).

In this event, could you model the guestimated reading by taking the linear average of the 24 hours of data prior that you do have, and extrapolating forward?
This would need to be displayed on the graph in a different fashion to make it clear this is an extrapolation based on partial data, instead of the actual readings.

Hi, you're probably aware of the problem that the Iffley Lock measurement station is currently allegedly giving incorrect measurements because of debris obstructing the weir. However, I'm sceptical that Iffley would be dramatically different from what's happening at Osney and Sandford.

So, I wondered if we could look into a predictor for the Iffley readings based on the measurements from nearby locks, trained on the historical data we have (I'm guessing a simple linear regression might do well). And then use said prediction to give a new flow rate estimate.